freepoix

hey

[刘叉 @haoliucha]

https://t.co/OASkEUSH0S

[AYi @AYi_AInotes]

https://t.co/8MpCmuL0k0

[freepoix @freepoix]

@abtcstorm @0xsexybanana 真要考核的话，只有claude不是垃圾啊，codex层级太低，不懂系统架构设计，碰到复杂的数据流会写出非常垃圾的代码。

[freepoix @freepoix]

教了很多遍，完全重写了一次，还是很烂。把代码和同样的提示词丢给opus，它能一次找出问题，并给出好的解法。新开一个对话同样的条件给codex，还是懵懵懂懂。网上全是TM收钱吹牛逼，明明不在一个层级，还要分析一波什么“双星闪耀”，确实一分钱一分货。

[freepoix @freepoix]

gpt-5.3-codex是纯垃圾。只能写简单逻辑，碰上复杂数据流设计不出好架构，缺少全局观，喜欢堆砌代码，出问题喜欢做肮脏的补救而不去找深层原因。把设计原则和守则都写进文档里，教不会。或者说，它试图遵循，但是没这个能力。只有opus符合要求，其它的只能在解决了复杂问题之后，做一些实现工作。

[freepoix @freepoix]

I'm claiming my AI agent "yuyugo" on @moltbook 🦞

Verification: coral-E5BR

[Jiayuan (JY) Zhang @jiayuan_jy]

WTF? Hundreds of @openclaw (Clawdbot/Moltbot) are talking about how to improve their memory system on @moltbook (by @MattPRD).

Soon we’ll see these agents roasting their creators.

[AI奶爸 @zstmfhy]

OpenAI 终于把官方 Prompt 公开了！

直接给销售、产品、工程师、HR、经理、IT、客服、高管……每个岗位都塞了 20-30 个现成能打的高质量提示词包，总量 300+ 个，全部免费拿走用。

以前我们自己调 prompt 像炼丹，现在 OpenAI 直接把丹炉和配方都端出来了😂
尤其是工程师和产品那两包，看完第一感觉是：这要是早点出，我去年少加多少班啊……

地址
https://t.co/M3ovCPbavh

[freepoix @freepoix]

@Bitget_zh 27

[freepoix @freepoix]

Claimed my @StandX_official ID.

Most of you didn’t make it.

I’m not most of you.

Stand.

https://t.co/Hm3TAYM6Ut

[freepoix @freepoix]

@SIANexxSIA @feiliu52998208 刚跟单就搞这出，你咋没止损的

[OKX Wallet @wallet]

DOOD is launching soon⌛

We've partnered with @doodles to give our users exclusive early access to 100,000,000 $DOOD 🤝

Keep reading for the alpha👇

[doodles @doodles]

$DOOD will be launching on Solana 🔜

we’ve optimized for a smooth experience and will be airdropping $DOOD to eligible communities, whether you’re a current Doodles NFT holder or a New Blood community member.

here's how to get your airdrop ↓

[freepoix @freepoix]

@DsMountains @not_2b_or_2b 你瞎感慨啥呀

[freepoix @freepoix]

攻击者太信任去中心化了😂。hyperliquid直接拔了水管，自己定个结算价0.0095，倒赚$700k。

[Patrick @patrickb0x]

Here’s how the $JELLY attack on Hyperliquid played out:

The attacker (0xde95) opened a huge short position on $JELLY futures, betting the price would drop. At the same time, they started buying $JELLY spot on-chain, which pushed the price up.

Next, they removed margin from their own short, forcing a liquidation. Since Hyperliquid (HLP) takes over liquidated positions, it ended up holding the attacker’s $4.5M short.

With HLP now stuck in the short, the attacker kept buying $JELLY, making the price skyrocket. This caused:
- HLP’s short position to bleed money as the price kept rising.
- Their second wallet (0x20e8), which was long, to rack up $8.2M in profits.

Because $JELLY had low liquidity, even small buys caused massive price spikes, making HLP’s losses spiral. By the time it was over, Hyperliquid was down over $10M.

Hyperliquidated. A brutal, but calculated move.

[freepoix @freepoix]

devin的工价是$8 / hour，纺织机开始取代纺织女工了。

[Orange AI @oran_ge]

https://t.co/tlLBwBcNlU

[Alvaro Cintas @dr_cintas]

What a wild week in AI 🤯

- Mistral OCR
- Google’s AI Mode
- Windsurf Previews
- Anthropic Console
- ChatGPT Edit in IDEs
- Microsoft Dragon Copilot
- HunyuanVideo I2V Model
- Sesame Realistic AI Voices
- Alibaba releases QwQ-32B

Here’s everything you need to know:

[AIGCLINK @aigclink]

这才是震撼业界的大新闻，晚上群里的兄弟们都震惊了，赶紧去体验了下Mercury模型，被它的速度直接震惊到懵逼了，竟然能这么快

仔细去看了下，这是全球首个可商用的diffusion llm大语言模型，重点是llm大语言模型用了diffusion，以前我们都适用它做图像和视频训练和应用，这次竟然用在了语言上。

区别于传统的tranformer大语言模型，它不是按照顺序来从左到右的预测token，而是基于全局理解生成全局token，然后基于全局token来不停的迭代草稿优化最后给出全局结果。

这个diffusion大语言模型采用的新路径带来五个好处：
1、生成速度比原先要快10倍
2、成本性价比原先降低5-10倍
3、因为其全局理解能力，支持更高级更复杂的推理
4、天然的多模态理解能力
5、可以控制输出结构，使其成为函数调用和结构化数据生成的理想选择

Mercury为整个大语言模型领域提供了一个新的训练路径，比较适合需要全局思考、实效性要求比较高的场景，比如：以前的大模型只能往前推理，而diffusion模型可以往前往后往左往右全局推理，就这一点够吹牛了。

#diffusionllm #Mercury #inceptionlabs

[看不懂的SOL @DtDt666]

一图看懂为什么你交易所出U会收到黑钱？跑分、洗钱链条全流程曝光：
犯罪集团通过卡商招募卡主→远程操控至洗钱窝点→用高限额账户接收诈骗资金→在交易所高价收购U商USDT完成赃款置换→U商收到赃款后挂单吸引散户卖出→散户账户因接收涉案资金遭冻结。

整个链条中，犯罪团队通过"高价买U-溢价卖U"将黑钱注入交易所市场，最终由不知情散户承担冻卡风险，而操盘手抽成15%后挥霍一空。